Anthropic Confirms Claude Mythos Testing After Draft Leak
San Francisco, US
Priyanshu Singh/Reuters
Source Analysis
What Happened
Why It Matters
What Happened
Why It Matters
Where Sources Agree
- arrows_inputAnthropic Confirms Mythos Leak: Coverage documents the leak of Anthropic's unreleased AI model, 'Claude Mythos,' from a draft blog post was caused by human error in a publicly accessible content management system, according to Anthropic's confirmation and Fortune reports.
- arrows_inputMythos Model Capabilities: Coverage broadly notes that Mythos is Anthropic's most powerful AI model, representing a 'step change' in capabilities and excelling in reasoning, coding, and cybersecurity, according to Anthropic's internal drafts and spokesperson.
- arrows_inputMythos Cybersecurity Risks: Reporting largely establishes that Anthropic's Mythos model poses unprecedented cybersecurity risks, with large-scale cyberattacks becoming more likely by 2026, according to Anthropic's draft blog post and expert warnings.
Where Sources Disagree
- arrows_outputData Exposure Nature: Anthropic attributed the data exposure to human error in its content management system, clarifying it was not a security breach. However, some outlets characterized the incident as an embarrassing security lapse, noting the publicly-searchable data store.
- arrows_outputAnthropic's AI Risk Motives: Anthropic officials privately warn government officials about AI risks and refuse military use; however, some outlets suggest the company is hyping dangers for strategic gain or to boost its IPO.
Timeline
March 31, 2026
Claude Code Source Leak Emerges: A separate but related incident on March 31 exposed source-map files that allowed near-complete reconstruction of Claude Code's internal TypeScript code; Anthropic said this was a release-packaging error (not a breach), that no sensitive customer data were exposed, and pledged fixes.
March 27, 2026
Cybersecurity Stocks Sell Off: Markets reacted to the revelations and risk warnings: major cybersecurity stocks tumbled (examples include Palo Alto Networks, Okta and CrowdStrike), with investors repricing the perceived threat to incumbent security vendors.
March 27, 2026
Anthropic Confirms and Acknowledges: Following the leak, Anthropic confirmed it is developing and testing the new model with early access customers, described it as a "step change" and the most capable model to date, and said the draft's exposure was caused by human error in a CMS configuration.
Summary by Ground AI
Sources
See All 62A varied selection of sources chosen by Ground to reflect the diversity of this story’s coverage.
Most Written About
The stories getting the most coverage from the last 24 hours
Source Analysis
Timeline
March 31, 2026
Claude Code Source Leak Emerges: A separate but related incident on March 31 exposed source-map files that allowed near-complete reconstruction of Claude Code's internal TypeScript code; Anthropic said this was a release-packaging error (not a breach), that no sensitive customer data were exposed, and pledged fixes.
March 27, 2026
Cybersecurity Stocks Sell Off: Markets reacted to the revelations and risk warnings: major cybersecurity stocks tumbled (examples include Palo Alto Networks, Okta and CrowdStrike), with investors repricing the perceived threat to incumbent security vendors.
March 27, 2026
Anthropic Confirms and Acknowledges: Following the leak, Anthropic confirmed it is developing and testing the new model with early access customers, described it as a "step change" and the most capable model to date, and said the draft's exposure was caused by human error in a CMS configuration.
